OpenAI published its own usage data across its enterprise customer base this month, and one number stands out from the rest: frontier firms, the top 10% of enterprise AI users by monthly usage, now generate 8.3 times as many output tokens per active user as typical firms. In January, that gap was 2.6x. It more than tripled in five months. That's not a story about adoption catching up. It's the opposite.
The number that matters
Most conversations about AI adoption focus on access: how many employees have a license, how many teams have rolled a pilot out. OpenAI's data measures something more useful, usage intensity per active user, and it shows the gap between leaders and laggards isn't closing as more companies get access. It's widening, and widening faster than it was five months ago. A typical firm waiting to catch up isn't chasing a fixed target. It's chasing one that's accelerating.
Where the growth is actually happening
The function-level detail is the part worth paying attention to. Since February, legal department AI usage at frontier firms grew 108x. Sales and recruiting each grew 41x. Engineering, the function most people assume is leading the way, grew a comparatively modest 5x. Plugin adoption sits at 21% of weekly active users at frontier firms versus 9% at typical firms, and skills adoption at 19% versus 3%. The growth isn't concentrated where the AI adoption conversation usually points. It's showing up in back-office and revenue functions that don't get the same attention in enterprise AI strategy decks.
From assistance to execution
There's a second signal in the same dataset. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. That's a real shift from asking AI questions to delegating it actual work, agentic execution rather than conversational assistance. Frontier firms aren't just using AI more. They're using it differently, handing over more substantive, delegated tasks instead of treating it as a smarter search bar.
The catch: this is OpenAI's own homework
Worth saying plainly: this data comes from OpenAI's own usage telemetry across its own enterprise customer base. It isn't an independently audited study, and it can't speak to companies building on other vendors' models. What it does show, credibly, is how usage intensity has evolved among a very large set of real paying customers over a defined six-month window. That's a meaningful signal about the direction of travel, even with the caveat that it's one company's view of its own product.
What this means for closing the gap
- Measure usage intensity per active user, not just how many people have access. Access without depth is where most enterprises are stuck, and it's invisible on a simple adoption-rate dashboard.
- Look at which functions are actually scaling usage, not just where you assumed it would happen. Legal outgrowing engineering by more than 20x is exactly the kind of signal a top-down rollout plan would miss.
- Treat "keeping pace" as a moving target, not a fixed bar. A gap that triples in five months compounds faster than most planning cycles can react to.
Where TrueHorizon fits
We help enterprises turn AI access into AI usage that actually compounds, the difference between a license nobody opens and a workflow that gets faster every quarter. That's not a checklist we're learning on your project. It's the expertise we bring to it. The gap OpenAI's data describes is real, and it rewards the companies that treat AI adoption as an operating discipline instead of a one-time rollout.
If you want to know where your organization actually stands against that gap, take our AI readiness assessment before it gets wider.
Written by
Deepankar Bhadrasen
Founding Engineer
Deepankar is an AI automation specialist and Founding Engineer at TrueHorizon AI, where he builds practical AI systems that help businesses streamline operations, reduce costs, and scale efficiently. He focuses on integrating custom AI agents and workflows with existing tools so teams can grow without expanding headcount.
